[Paper Review] Network assisted analysis to reveal the genetic basis of autism

Li Liu, Jing Lei+1 more|arXiv (Cornell University)|Jun 2, 2015

Genomic variations and chromosomal abnormalities62 references35 citations

TL;DR

This paper proposes DAWN, a network-assisted statistical framework that integrates gene co-expression networks with genetic association scores to identify autism spectrum disorder (ASD) risk genes. By using a novel partial neighborhood selection (PNS) algorithm to estimate gene networks focused on potentially disease-relevant regions and combining this with a hidden Markov random field (HMRF) model, the method improves detection power in high-dimensional settings. The approach identified 333 high-confidence ASD risk genes, including novel candidates, by leveraging brain-specific developmental gene expression data and whole-exome sequencing data from over 16,000 samples.

ABSTRACT

While studies show that autism is highly heritable, the nature of the genetic basis of this disorder remains illusive. Based on the idea that highly correlated genes are functionally interrelated and more likely to affect risk, we develop a novel statistical tool to find more potentially autism risk genes by combining the genetic association scores with gene co-expression in specific brain regions and periods of development. The gene dependence network is estimated using a novel partial neighborhood selection (PNS) algorithm, where node specific properties are incorporated into network estimation for improved statistical and computational efficiency. Then we adopt a hidden Markov random field (HMRF) model to combine the estimated network and the genetic association scores in a systematic manner. The proposed modeling framework can be naturally extended to incorporate additional structural information concerning the dependence between genes. Using currently available genetic association data from whole exome sequencing studies and brain gene expression levels, the proposed algorithm successfully identified 333 genes that plausibly affect autism risk.

Motivation & Objective

• To overcome the challenge of weak, dispersed genetic signals in ASD gene discovery by integrating functional genomics data.
• To improve statistical power in high-dimensional genetic association studies by focusing network estimation on biologically relevant regions of the gene network.
• To develop a flexible, scalable framework that incorporates additional biological network information (e.g., transcription factor targets) to enhance risk gene detection.
• To identify a robust, biologically interpretable set of ASD risk genes beyond those detected by marginal association tests alone.

Proposed method

• Proposes a partial neighborhood selection (PNS) algorithm that incorporates node-specific genetic association scores into network estimation to improve accuracy and computational efficiency.
• Uses a hidden Markov random field (HMRF) model to combine estimated gene networks with individual gene-level association statistics (e.g., TADA p-values) in a unified probabilistic framework.
• Extends the HMRF model to include additional structural information, such as transcription factor (e.g., FMRP) target genes, via an Ising model formulation.
• Employs a tuning parameter selection strategy to balance model smoothness and signal retention, with cross-validation used to select optimal regularization.
• Applies Bayesian false discovery rate (FDR) procedures to the HMRF output to identify significant risk genes with controlled error rates.
• Validates the method using simulated data and applies it to real-world ASD data from whole-exome sequencing and BrainSpan transcriptome data.

Experimental results

Research questions

• RQ1 Can integrating gene co-expression networks with genetic association scores improve the detection of autism risk genes in high-dimensional genomic data?
• RQ2 Does focusing network estimation on regions surrounding potentially disease-associated genes enhance statistical power and reduce false positives?
• RQ3 How does the inclusion of additional biological network information (e.g., FMRP targets) improve risk gene discovery?
• RQ4 Can the proposed method identify novel ASD risk genes beyond those detected by marginal association tests?

Key findings

• The DAWN framework identified 333 high-confidence autism risk genes using whole-exome sequencing and brain-specific gene expression data.
• The method successfully detected 118 genes with at least one de novo loss-of-function (dnLoF) mutation, including four novel genes (TRIP12, RIMBP2, ZNF462, ZNF238) when FMRP target information was incorporated.
• The PNS-based network estimation outperformed standard methods like Glasso in terms of power and precision, particularly in high-dimensional settings with limited sample sizes.
• The model maintained high consistency across tuning parameters, identifying nearly all 95 high-confidence genes from the TADA FDR < 0.3 list regardless of smoothing level.
• Incorporating FMRP target data significantly improved detection power, with a statistically significant improvement (p < 0.005) in the Ising model.
• The intersection of results across multiple tuning parameters yielded a robust and reliable list of candidate ASD risk genes, minimizing false discovery.